Multi-band multipurpose spectrum filtering system

Abstract

The invention belongs to the technical field of spectrum filtering, and discloses a multiband multipurpose spectrum filtering system. Comprising a data acquisition module used for acquiring spectral data of an experimental sample under a preset light source; the missing value processing module is used for performing missing value processing on the spectral data to obtain complete spectral data; the abnormal value detection module is used for judging an abnormal value in the complete spectral data and carrying out abnormal value processing on the complete spectral data to obtain normal spectral data; the filtering module comprises a wavelet de-noising unit and a baseline filtering unit, and the wavelet de-noising unit is used for performing de-noising processing on the normal spectral data to obtain filtered spectral data; the baseline filtering unit is used for performing baseline correction on the filtered spectral data to obtain accurate spectral data; the obtained accurate spectral data can be directly applied to the process flow of the subsequent optical field, and the working efficiency can be remarkably improved by using the high-quality data.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral filtering, and more specifically, to a multi-band and multi-purpose spectral filtering system. Background Art

[0002] The patent with application publication number CN107203055A discloses a light filtering method for a spectral analysis system, which determines the spectral range to be analyzed, that is, the wavelength range of the light entering the spectral analysis system; according to the above spectral range, determines the specific parameters such as the acoustic wave frequency of the acousto-optic tunable filter; adjusts the positions of the plane mirror I and the plane mirror II so that the first filtered light enters the acousto-optic tunable filter again; adjusts the position of the detector so that it can collect the second filtered light; adjusts the angle between the acousto-optic tunable filter and the incident light, and finds a suitable δ value, that is, the difference between the center wavelengths of the first filtered light and the second filtered light, so as to improve the spectral resolution while ensuring that the peak of the second filtered light at the center frequency of the spectrum is strong enough. A single acousto-optic tunable filter is used, the structure is simplified, the loss of extraordinary incident light can be minimized, the spectral resolution is increased and the side lobes are suppressed.

[0003] In the prior art, most of the methods for processing missing values ​​of spectral data are not accurate enough and are not effective for complex spectral data. For example, a linear interpolation algorithm is used to process missing values ​​of spectral data. In terms of anomaly detection, the prior art usually uses a fixed threshold to judge outliers. This method is not suitable for processing multi-dimensional spectral data sets. If the scale of the spectral data set is large, the use of traditional outlier detection methods is likely to lead to excessive computational complexity, affecting the efficiency of the algorithm. The prior art often uses filters to process noise in spectral data, but the performance and parameters of the filter are likely to affect the effect of noise processing. If the filter performance is poor, the noise processing efficiency will be low. For the baseline offset problem of spectral data, the traditional baseline correction method used in the prior art is difficult to process spectral data. If the parameter values ​​are inappropriate, it is easy to cause overfitting or underfitting, which ultimately affects the quality of the spectral data.

[0004] In view of this, the present invention proposes a multi-band and multi-purpose spectral filtering system to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: a multi-band multi-purpose spectral filtering system, comprising: A data acquisition module is used to collect spectral data of experimental samples under a preset light source; Missing value processing module, used to process missing values ​​of spectral data to obtain complete spectral data; An outlier detection module is used to determine outliers in the complete spectral data and perform outlier processing on the complete spectral data to obtain normal spectral data; The filtering module includes a wavelet denoising unit and a baseline filtering unit. The wavelet denoising unit is used to perform denoising on normal spectral data to obtain filtered spectral data; the baseline filtering unit is used to perform baseline correction on the filtered spectral data to obtain accurate spectral data; and each module is connected by wire and / or wireless means.

[0006] Furthermore, spectral data is collected using a preset light collection method; the spectral data includes wavelength, light intensity and a corresponding timestamp.

[0007] Furthermore, the method of processing missing values ​​of spectral data includes: The Monte Carlo algorithm is used to process the wavelength missing points of the spectral data, and the numerical values ​​generated by the algorithm are used to fill the wavelength missing values ​​of the spectral data. The effect of the wavelength missing value processing is evaluated by calculating the mean square error of the spectral data after the wavelength missing value processing; when the mean square error is less than the preset mean square error threshold, the wavelength missing value processing effect is optimal, and a wavelength filled data set is obtained at this time; Construct a machine learning model and use the RNN recurrent neural network model as the basic framework of the machine learning model; use the machine learning model to process the missing values ​​of light intensity in the wavelength filling dataset, and use the predicted values ​​of the machine learning model to fill the missing values ​​of light intensity in the wavelength filling dataset; Collect historical complete spectral data as training labels for machine learning models; randomly select data points, and mark the timestamps of these data points to obtain marked data points; delete the light intensity values ​​of all marked data points to obtain a missing light intensity data set, and use the missing light intensity data set as a training set for the machine learning model; Construct a custom scoring mechanism to calculate the custom score of each data point in the missing light intensity dataset and the average of the custom scores of all data points; define a machine learning loss function and calculate the function value of the machine learning loss function based on the custom score of each spectral data point and the average of the custom scores of all data points.

[0008] Furthermore, the calculation formula for the custom score of each data point in the missing light intensity dataset is: ;in, Indicates the missing light intensity data set Custom scoring for each data point; Indicates the missing light intensity data set Physical law scores for each data point; represents the weight of the physical law score; Indicates the missing light intensity data set Smoothed scores of data points; represents the weight of the smoothing score; Indicates the missing light intensity data set missing scores for data points; represents the weight of missing ratings; Physics Law Rating ;in, Indicates the missing light intensity data set The predicted value of light intensity for each data point; Indicates the missing light intensity data set The true value of the light intensity at each data point; represents the standard deviation of light intensity; Smoothed Scoring ;in, Represents the total number of data points in the missing light intensity dataset; Missing rating ;in, Indicates the number of missing light intensity data sets in the preset The weight of the light intensity of each data point; Indicates the missing light intensity data set The light intensity value of each data point; Machine Learning Loss Functions ;in Represents the mean of the custom scores for all data points in the missing light intensity dataset; Training machine learning models using missing light intensity datasets; weighting physical law scores using gradient descent , weight of smoothing score and weights for missing ratings Update; when the function value of the machine learning loss function no longer decreases, fix the parameters to obtain a trained machine learning model.

[0009] Furthermore, the method of performing outlier processing on the complete spectral data includes: Based on the timestamp of each complete spectral data point in the complete spectral data, the complete spectral data is sorted in chronological order to obtain ordered complete spectral data; the ordered complete spectral data is standardized to obtain standardized spectral data; characteristic coefficients are defined, and anomaly thresholds are calculated based on the characteristic coefficients; the standardized spectral data are processed using an anomaly detection algorithm to obtain an anomaly score for each standardized spectral data point in the standardized spectral data; all standardized spectral data points in the standardized spectral data whose anomaly scores are greater than the anomaly threshold are selected to form an anomaly data set; Calculate the information entropy of each abnormal spectral data point in the abnormal data set, calculate the weight vector of each abnormal spectral data point in the abnormal data set based on the information entropy of all abnormal spectral data points; calculate the abnormal factor score of each abnormal spectral data point in the abnormal data set; calculate the final abnormal value score based on the abnormal value score, weight vector and abnormal factor score of each abnormal spectral data point in the abnormal data set; if the final abnormal value score of any abnormal spectral data point in the abnormal data set is greater than the preset final abnormal value score threshold, the abnormal spectral data point is determined to be an error spectral data point; delete all error spectral data points to obtain normal spectral data.

[0010] Furthermore, the method of calculating the abnormal threshold includes: Characteristic coefficient ;in, represents the number of standardized spectral data points in the standardized spectral data; Indicates The wavelength value of the standardized spectral data point; Represents the wavelength mean of all standardized spectral data points in the standardized spectral data; Indicates The light intensity value of the standardized spectral data point; Represents the mean light intensity of all standardized spectral data points in the standardized spectral data; Normalize the characteristic coefficients of all standardized spectral data points to obtain the characteristic coefficient vector ; The characteristic coefficient vector All characteristic coefficients in are arranged in descending order from large to small, and the top characteristic coefficients to form a characteristic coefficient set; summing all characteristic coefficients in the characteristic coefficient set, and calculating an abnormal threshold based on the sum of all characteristic coefficients in the characteristic coefficient set; Abnormal Threshold ;in, Represents the sum of all characteristic coefficients in the characteristic coefficient set; represents the regulating factor; The calculation formula of the weight vector is: ;in, Indicates abnormal spectral data points The weight vector of Represents the number of abnormal spectral data points in the abnormal data set; Indicates abnormal spectral data points Information entropy of represents an adaptive function; ;in, Indicates abnormal spectral data points The distribution probability of wavelength; Indicates abnormal spectral data points The distribution probability of light intensity; Adaptive function ;in represents an adaptive adjustment parameter; Indicates abnormal spectral data points The wavelength value of Indicates abnormal spectral data points The light intensity value; Represents the maximum value of wavelength among all abnormal spectral data points; Represents the maximum value of light intensity among all abnormal spectral data points.

[0011] Furthermore, the method of calculating the abnormal factor score includes: For any two abnormal spectral data points in the abnormal data set, the Euclidean distance between them is calculated, and a distance matrix is ​​constructed based on the Euclidean distance between any two abnormal spectral data points; the smallest value is selected from each row of the distance matrix. The minimum Euclidean distance and sum of all abnormal spectral data points are summed to obtain the distance vector ; where the distance vector Any element in represents the minimum Euclidean distance sum of any abnormal spectrum data point; Represents the transpose of a vector; The inverse of the minimum Euclidean distance sum of any abnormal spectral data point is taken to obtain the abnormal factor score of the abnormal spectral data point; Final outlier score ;in, Indicates abnormal spectral data points outlier score; Indicates abnormal spectral data points The abnormal factor score of Represents a fusion weight.

[0012] Furthermore, the method of performing denoising on the normal spectral data includes: Preset wavelet basis function, from less than or equal to A positive integer is randomly selected from the positive integers as the decomposition layer number of the wavelet decomposition; the normal spectral data is subjected to wavelet decomposition based on a preset wavelet basis function to obtain detail coefficients; the detail coefficients are threshold processed based on a preset detail coefficient threshold to obtain denoised detail coefficients; the denoised detail coefficients are signal reconstructed using an inverse wavelet transform to obtain filtered spectral data.

[0013] Furthermore, the method of performing baseline correction on the filtered spectral data includes: Selecting every five consecutive filtered spectrum data points from the filtered spectrum data as a group of correction points; taking the midpoint of any two adjacent correction points in a group of correction points, dividing the wavelength into three wavelength sub-intervals based on any two adjacent midpoints, constructing a polynomial function based on the wavelength sub-intervals; calculating fitting values ​​based on the polynomial function of each wavelength sub-interval; updating the filtered spectrum data based on the fitting values ​​to obtain updated accurate spectrum data;

[0014] The calculation formula of the polynomial function is: ;in, Indicates The polynomial function value corresponding to the wavelength value of any filtered spectrum data point in the wavelength sub-intervals; Indicates The wavelength value of any filtered spectrum data point in the wavelength sub-intervals; , , and are the polynomial coefficients; Construct an objective function, optimize the dynamic parameters of the objective function, and obtain the optimal dynamic parameters; calculate the optimal polynomial coefficient vector based on the optimal dynamic parameters; the elements in the optimal polynomial coefficient vector are the polynomial coefficients; Objective Function ;in, Indicates the number of filtered spectrum data points in any wavelength subinterval in the filtered spectrum data; Indicates the number of The polynomial function value corresponding to the wavelength value of the filtered spectrum data point; express The corresponding wavelength sub-interval The light intensity value of each filtered spectral data point; Indicates a dynamic parameter.

[0015] Furthermore, the method of optimizing the dynamic parameters of the objective function includes: Preset the initial value of the dynamic parameter, and calculate the dynamic norm based on the initial value of the dynamic parameter; continuously adjust the initial value of the dynamic parameter until the dynamic norm reaches the minimum value, and fix the dynamic parameter at this time as the optimal dynamic parameter; Dynamic norm ;in, Represents the light intensity vector composed of the light intensity values ​​of all filtered spectral data points in the wavelength sub-interval corresponding to the objective function; represents a polynomial matrix; Indicates Dynamic parameters of the secondary adjustment; Polynomial Matrix ;in, Represents the wavelength value of the first filtered spectrum data point in the wavelength subinterval corresponding to the objective function; Indicates the first wavelength subinterval corresponding to the objective function The wavelength value of the filtered spectral data point; Optimal polynomial coefficient vector ;in, represents the optimal dynamic parameter; represents an identity matrix; represents the transpose of a polynomial matrix; The calculation formula for updating the filtered spectrum data is: ; Represents any wavelength value in precise spectral data The corresponding light intensity value; Indicates wavelength value The corresponding fitted values; Indicates wavelength value The corresponding light intensity value in the filtered spectral data; Indicates wavelength value The maximum difference between the light intensity value of the wavelength sub-interval in the corresponding filtered spectrum data and the fitted value; Represents a constant.

[0016] The technical effects and advantages of the multi-band multi-purpose spectral filtering system of the present invention are as follows: By collecting spectral data, the missing value processing module, outlier detection module and filtering module are used to finely process the collected spectral data, realizing a multi-band and multi-purpose spectral filtering system; compared with existing experience, the missing values ​​in the spectral data are processed using interpolation algorithms and machine learning models, which not only improves the accuracy of the predicted values, but also captures the important features in the spectral data; at the same time, a custom scoring mechanism is introduced to calculate the loss function, so that the model can better adapt to the characteristics and laws of spectral data; in terms of outlier processing, an efficient outlier judgment method is adopted. Compared with the traditional fixed threshold, the use of dynamic threshold to judge outliers is more flexible and accurate; the spectral data is denoised by wavelet transform, and the spectral data is baseline corrected by cubic interpolation method, which effectively removes the background noise and interference signals in the spectral data, making the spectral data more accurate and indirectly improving the system work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a multi-band and multi-purpose spectral filtering system of the present invention; Figure 2 It is a schematic diagram of a multi-band and multi-purpose spectral filtering method of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Embodiment 1; See also Figure 1 As shown, the multi-band multi-purpose spectral filtering system described in this embodiment includes: A data acquisition module is used to collect spectral data of experimental samples under a preset light source; Missing value processing module, used to process missing values ​​of spectral data to obtain complete spectral data; An outlier detection module is used to determine outliers in the complete spectral data and perform outlier processing on the complete spectral data to obtain normal spectral data; The filtering module includes a wavelet denoising unit and a baseline filtering unit. The wavelet denoising unit is used to perform denoising on normal spectral data to obtain filtered spectral data; the baseline filtering unit is used to perform baseline correction on the filtered spectral data to obtain accurate spectral data; the accurate spectral data is applied to subsequent process flows in the optical field; and each module is connected by wire and / or wireless means.

[0020] Spectral data is collected using a preset light collection method; the light collection method includes taking a certain chemical liquid solvent as a sample, using a white light source to provide energy to make the sample generate a spectral signal (because white light is a mixture of multiple wavelengths, it covers a wide wavelength range from ultraviolet to infrared and can provide more comprehensive spectral information), selecting filters of different specifications and placing them between the white light source and the sample (filters of different specifications refer to filters that process different wavelength ranges, such as ultraviolet filters used to process spectral signals within the ultraviolet wavelength range), ensuring that each filter is at an equal distance from the sample (ensuring that the distance between the filter and the sample is equal is to eliminate the error of different spectral signal intensities caused by different distances between each filter and the sample); using a spectrometer (such as a Raman spectrometer) to receive the spectral signal of each filter to obtain spectral data; the spectral data includes wavelength, light intensity and corresponding timestamp; because the spectral data contains a time dimension, each data point in the spectral data reflects the wavelength and light intensity of the spectrum at the current moment, and records the spectral changes of the sample at different times.

[0021] In the process of collecting spectral data, due to equipment factors or environmental interference, the collected spectral data may contain missing values ​​and outliers. Therefore, the purpose of exception processing of spectral data is to fill in missing values ​​and correct outliers. Equipment factors include: failure of white light source, resulting in insufficient light intensity of white light or complete failure, resulting in missing spectral data; sensor in the spectrometer has reduced sensitivity or damage, and cannot accurately receive spectral signals, resulting in missing spectral data; environmental interference includes: temperature fluctuations in the experimental environment affect the performance of the spectrometer. For example, the spectrometer is equipped with a temperature-sensitive sensor, and temperature changes cause the spectral data of certain wavelengths to be lost or abnormally enhanced; if in a high humidity environment, the optical components inside the spectrometer (such as lenses and reflectors) are prone to fogging, affecting the transmission and reception of spectral signals.

[0022] The methods for handling missing values ​​of spectral data include: The Monte Carlo algorithm is used to process the wavelength missing points of the spectral data (the wavelength missing points refer to the data points in the spectral data that lack wavelength values), and the numerical values ​​generated by the algorithm are used to fill the wavelength missing values ​​of the spectral data. The effect of the wavelength missing value processing is evaluated by calculating the mean square error of the spectral data after the wavelength missing value processing; when the mean square error is less than the preset mean square error threshold, the wavelength missing value processing effect is optimal, and a wavelength-filled data set is obtained at this time; the Monte Carlo algorithm is a numerical calculation method based on random sampling, which determines the position of the wavelength missing point by scanning the spectral data, and uses the Monte Carlo algorithm to generate random numbers to fill the wavelength values ​​of the wavelength missing points, and obtains several different wavelength-filled data sets after multiple rounds of filling; the filling effects of different wavelength-filled data sets are evaluated, and the wavelength-filled data set with the best filling effect is taken as the final wavelength missing value processing result.

[0023] Spectral data usually contains rich information, and there may be a complex nonlinear relationship between the light intensity of different data points. Using traditional missing value processing methods (such as linear interpolation and mean filling methods, etc.) to process the missing values ​​of light intensity of spectral data cannot achieve accurate and efficient results. Therefore, we choose to build a machine learning model to process the missing values ​​of light intensity of spectral data. The model accurately predicts the missing values ​​of light intensity by learning the spectral features of the spectral data in the training set.

[0024] Build a machine learning model and use the RNN recurrent neural network model as the basic framework of the machine learning model.

[0025] Collect historical complete spectral data as training labels for machine learning models; randomly select data points ( is less than the total number of spectral data points in the historical complete spectral data), and the timestamps of these data points are marked to obtain marked data points; the light intensity values ​​of all marked data points are deleted to obtain a missing light intensity data set, and the missing light intensity data set is used as a training set for the machine learning model.

[0026] The RNN recurrent neural network model performs forward propagation to obtain the predicted value of the missing value of light intensity in the missing light intensity data set, and calculates the function value of the traditional loss function (such as the signal-to-noise ratio function) based on the predicted value of light intensity and the true value of light intensity (the light intensity value of the data point corresponding to the timestamp of the marked data point in the historical complete spectral data). When the function value of the traditional loss function no longer decreases, the trained RNN recurrent neural network model is obtained; for spectral type data, it usually follows certain physical or chemical laws, and only using traditional loss functions is likely to ignore the laws in the data, causing the model to generate prediction results that do not conform to the laws. For example, in infrared spectral analysis, the light intensity at certain specific wavelengths will increase with the increase in the concentration of the test sample, and only using traditional loss functions cannot reflect the change in light intensity; the missing values ​​of spectral data may be concentrated in certain specific light intensity areas, and the traditional loss function cannot distinguish the importance of different light intensities, resulting in the model being ineffective when processing missing values; therefore, a custom scoring mechanism is introduced, and the function value of the loss function is calculated based on the custom scoring.

[0027] A custom scoring mechanism is constructed to calculate the custom score of each data point in the missing spectral dataset and the average of the custom scores of all data points; a machine learning loss function is defined to calculate the function value of the machine learning loss function based on the custom score of each data point and the average of the custom scores of all data points.

[0028] The custom score for each data point in the missing spectrum dataset is calculated as: ;in, Indicates the missing spectral data set Custom scoring for data points ( less than or equal to the total number of data points in the missing spectrum dataset); Indicates the missing spectral data set Physical law scores for each data point; represents the weight of the physical law score; Indicates the missing spectral data set Smoothed scores of data points; represents the weight of the smoothing score; Indicates the missing spectral data set missing scores for data points; Indicates the weight of the missing score.

[0029] Physics Law Rating ;in, Indicates the missing spectral data set The predicted value of light intensity for each data point; Indicates the missing spectral data set The true value of the light intensity at each data point; It represents the standard deviation of light intensity (obtained by calculating the square root of the variance of light intensity in the historical complete spectral data); the physical law score is used to reflect the difference between the predicted value and the true value. The smaller the score, the closer the predicted value is to the true value.

[0030] Smoothed Scoring ;in, Indicates the total number of data points in the missing spectral data set; spectral data is generally smooth, that is, the change in the light intensity of adjacent data points is small. The smaller the score value, the smoother the change in the spectral data.

[0031] Missing rating ;in, Indicates the missing spectral data set. The weight of the light intensity of each data point; Indicates the missing spectral data set The light intensity value of the data point (if the light intensity value is missing, it defaults to ); Using preset weights to distinguish the importance of different light intensities can make the model pay more attention to areas with more missing values. For example, the proportion of missing values ​​of light intensity of data points in a certain wavelength is relatively high, so a larger weight is given to the light intensity of the data points in this wavelength. By adjusting the weights, the model can handle missing values ​​more accurately.

[0032] Machine Learning Loss Functions ;in Represents the mean of the custom scores for all data points in the missing spectra dataset.

[0033] The introduction of a custom scoring mechanism to calculate the machine learning loss function further enhances the flexibility of the machine learning model, which can help the machine learning model better capture the important features in the spectral data and improve the accuracy of the machine learning model's predictions.

[0034] Training machine learning models using missing spectral datasets; weighting physical law scores using gradient descent , weight of smoothing score and weights for missing ratings Update; when the function value of the machine learning loss function no longer decreases, fix the parameters to obtain a trained machine learning model.

[0035] The trained machine learning model is used to process missing values ​​in the spectral data, and the predicted values ​​of the machine learning model are used to fill in the missing values ​​in the spectral data.

[0036] Using models to predict missing values ​​reduces the noise and errors introduced by missing values ​​and obtains more accurate prediction values. When dealing with situations with different data types and more complex data sets, using models to process data is more flexible and efficient.

[0037] The methods for handling outliers on complete spectral data include: Traditional anomaly detection usually uses a method based on a certain threshold to judge outliers, such as using quartiles to judge outliers. For multidimensional data sets such as complete spectral data, using a fixed threshold for judgment may not be able to accurately find outliers. Therefore, it is necessary to design an efficient outlier judgment method.

[0038] Based on the timestamp of each complete spectral data point in the complete spectral data, the complete spectral data are sorted in chronological order to obtain ordered complete spectral data; the ordered complete spectral data are standardized to obtain standardized spectral data; characteristic coefficients are defined, and anomaly thresholds are calculated based on the characteristic coefficients; the standardized spectral data are processed using an anomaly detection algorithm (such as an isolation forest algorithm) to obtain an outlier score for each standardized spectral data point in the standardized spectral data; the method of processing the standardized spectral data using the isolation forest algorithm includes: constructing all standardized spectral data points into a number of isolated trees; for each standardized spectral data point, calculating the average path length of the data point in all isolated trees (the shorter the path length, the easier it is for the data point to be isolated, and it is likely to be an anomaly point; the longer the path length, the harder it is for the data point to be isolated, and it is more likely to be a normal point); based on the average path length, the outlier score of each standardized spectral data point is calculated using the known outlier score formula of the isolation forest algorithm; and all standardized spectral data points in the standardized spectral data whose outlier scores are greater than the outlier threshold are selected to form an outlier data set.

[0039] Calculate the information entropy of each abnormal spectral data point in the abnormal data set, calculate the weight vector of each abnormal spectral data point in the abnormal data set based on the information entropy of all abnormal spectral data points; calculate the abnormal factor score of each abnormal spectral data point in the abnormal data set; calculate the final abnormal value score based on the abnormal value score, weight vector and abnormal factor score of each abnormal spectral data point in the abnormal data set; if the final abnormal value score of any abnormal spectral data point in the abnormal data set is greater than the preset final abnormal value score threshold, the abnormal spectral data point is determined to be an error spectral data point; delete all error spectral data points to obtain normal spectral data.

[0040] The methods for calculating anomaly thresholds include: A characteristic coefficient is introduced to quantify the differences in wavelength and intensity between different standardized spectral data points, and based on these differences, it can better reflect the extent to which the wavelength and intensity of some standardized spectral data points deviate from the normal range.

[0041] Characteristic coefficient ;in, represents the number of standardized spectral data points in the standardized spectral data; Indicates The wavelength value of the standardized spectral data point; Represents the wavelength mean of all standardized spectral data points in the standardized spectral data; Indicates The light intensity value of the standardized spectral data point; Represents the mean light intensity of all standardized spectral data points in the standardized spectral data.

[0042] Normalize the characteristic coefficients of all standardized spectral data points to obtain the characteristic coefficient vector .

[0043] The characteristic coefficient vector All characteristic coefficients in are arranged in descending order from large to small, and the top feature coefficients to form a feature coefficient set; all feature coefficients in the feature coefficient set are summed, and an abnormality threshold is calculated based on the sum of all feature coefficients in the feature coefficient set.

[0044] Abnormal Threshold ;in, Represents the sum of all characteristic coefficients in the characteristic coefficient set; represents the adjustment factor ( ).

[0045] The anomaly threshold is calculated based on the characteristic coefficient, and the suspected outliers in the standardized spectral data are preliminarily screened to better evaluate the degree of abnormality of the standardized spectral data points; in large-scale data sets, directly calculating the outlier scores of all standardized spectral data points will result in excessive calculations in a short period of time, affecting the efficiency of the algorithm; the anomaly threshold is adaptively adjusted according to specific situations, which improves the flexibility of the algorithm and enables the algorithm to have higher performance when dealing with different situations.

[0046] The calculation formula of the weight vector is: The weight vector of abnormal spectral data points is the collection of information entropies of all abnormal spectral data points. Any element in the vector corresponds to the information entropy of any abnormal spectral data point. However, during the experiment, the information entropy of some abnormal spectral data points may affect the wavelength value and light intensity value of the point due to environmental interference. Therefore, an adaptive function is added to the calculation formula of the weight vector to adjust the light intensity and wavelength of a certain abnormal spectral data point.

[0047] ;in, Indicates abnormal spectral data points The weight vector of Represents the number of abnormal spectral data points in the abnormal data set; Indicates abnormal spectral data points Information entropy of represents an adaptive function.

[0048] ;in, Indicates abnormal spectral data points The distribution probability of wavelength; Indicates abnormal spectral data points The distribution probability of light intensity.

[0049] Adaptive function ;in Represents an adaptive adjustment parameter ( is a variable constant, ); Indicates abnormal spectral data points The wavelength value of Indicates abnormal spectral data points The light intensity value; Represents the maximum value of wavelength among all abnormal spectral data points; Represents the maximum value of the light intensity among all abnormal spectral data points; based on the adaptive function, the weight vector of the abnormal spectral data points is dynamically adjusted instead of simply performing global statistics, which makes the algorithm more flexible, better captures the anomalies of local data, and improves the overall robustness.

[0050] The methods for calculating the outlier factor score include: Calculating the final outlier score of the abnormal spectral data point based only on the outlier score and weight vector is not rigorous enough. The outlier score and weight vector only consider the global degree of abnormality of the abnormal spectral data point. Therefore, an anomaly factor score is introduced to calculate the local degree of abnormality of the abnormal spectral data point.

[0051] For any two abnormal spectral data points in the abnormal data set, the Euclidean distance between them is calculated, and a distance matrix (the size of the matrix is ); for example, the distance matrix ; The first row represents abnormal spectral data points The Euclidean distance to any other abnormal spectral data point. The last row represents the abnormal spectral data point The Euclidean distance with any other abnormal spectral data point, and the same applies to the rest of the middle rows.

[0052] Select the smallest value from each row of the distance matrix The minimum Euclidean distance and sum of all abnormal spectral data points are summed to obtain the distance vector ; where the distance vector Any element in represents the minimum Euclidean distance sum of any abnormal spectrum data point; Represents the transpose of a vector.

[0053] The inverse of the minimum Euclidean distance sum of any abnormal spectral data point is taken to obtain the abnormal factor score of the abnormal spectral data point.

[0054] The anomaly factor score takes into account the Euclidean distance between any two anomaly spectral data points in the anomaly data set, and then captures the density difference between the local area composed of a certain anomaly spectral data point and its adjacent anomaly spectral data points and the local area composed of other anomaly spectral data points; the anomaly factor score is added to the final anomaly value score, so that the final anomaly value score comprehensively considers the global anomaly degree and the local anomaly degree, which can more accurately identify the true error spectral data points and improve the accuracy and efficiency of anomaly processing.

[0055] Final outlier score ;in, Indicates abnormal spectral data points outlier score; Indicates abnormal spectral data points The abnormal factor score of Represents a fusion weight ( , the fusion weight is used to adjust the importance of the final outlier score to the outlier score and the outlier factor score).

[0056] After missing value processing and outlier processing, the spectral data ensures the integrity and consistency of the data set; however, there are still some minor errors in the processed spectral data, such as systematic errors caused by experimental instruments or random noise caused by environmental factors; these minor errors can easily lead to reduced accuracy and low quality of the obtained spectral data, so these data need to be denoised.

[0057] The methods for denoising normal spectral data include: Wavelet transform is used to preprocess the normal spectral data to remove the noise, provide cleaner data for subsequent baseline correction, and indirectly improve the processing efficiency of normal spectral data.

[0058] Preset wavelet basis functions (such as function), from less than or equal to ( A positive integer is randomly selected from the positive integers (representing the number of normal spectrum data points in the normal spectrum data) as the decomposition layer number of the wavelet decomposition; the normal spectrum data is subjected to wavelet decomposition based on a preset wavelet basis function to obtain detail coefficients (the detail coefficient refers to the high-frequency random noise information in the normal spectrum data); the detail coefficients are subjected to threshold processing based on a preset detail coefficient threshold (using a soft threshold processing method, when the absolute value of the detail coefficient is less than the preset detail coefficient threshold, the detail coefficient is set to zero; when the absolute value of the detail coefficient is greater than the preset detail coefficient threshold, the detail coefficient is subtracted from the preset detail coefficient threshold; the soft threshold can make the signal smoother and reduce the risk of overfitting), and the denoised detail coefficients are obtained; the denoised detail coefficients are subjected to signal reconstruction using an inverse wavelet transform to obtain filtered spectrum data.

[0059] Compared with other denoising algorithms, using wavelet transform to process noise in normal spectral data is more flexible, and appropriate wavelet basis functions can be selected according to different data characteristics; wavelet transform can well preserve the edges and mutation parts in the signal, and will not directly filter and cause signal distortion; among the existing denoising algorithms, the methods and tools of wavelet transform are already quite mature and easy to implement.

[0060] Methods for baseline correction of filtered spectral data include: There are various background noises and interference signals in the unprocessed spectral lines, such as scattered light generated by the optical components inside the spectrometer, interference from other light sources during the experiment, electronic noise in the equipment itself, and changes in the composition or structure of the samples used in the experiment, which cause changes in the spectral signals; the background noise and interference signals are superimposed on the real spectral signals collected, causing the baseline of the spectrum to shift. Therefore, a cubic interpolation method is used to fit the new baseline to update the original spectral lines of the filtered spectral data to obtain more accurate spectral data.

[0061] Select every five consecutive filtered spectrum data points from the filtered spectrum data as a set of correction points; take the midpoint of any two adjacent correction points in the set of correction points, divide the wavelength into three wavelength sub-intervals based on any two adjacent midpoints, and construct a polynomial function based on the wavelength sub-intervals; for example , , , and is a set of calibration points, and The midpoint of , and The midpoint of , and The midpoint of , and The midpoint of ; Based on and The polynomial function constructed by the wavelength subintervals composed of the corresponding wavelengths is recorded as ,in, Indicated by and The wavelength value of any filtered spectrum data point in the wavelength sub-interval composed of the corresponding wavelengths; based on and The polynomial function constructed by the wavelength subintervals composed of the corresponding wavelengths is recorded as ,in, Indicated by and The wavelength value of any filtered spectrum data point in the wavelength sub-interval composed of the corresponding wavelengths; based on and The polynomial function constructed by the wavelength subintervals composed of the corresponding wavelengths is recorded as ;in, Indicated by and The wavelength value of any filtered spectrum data point in the wavelength subinterval composed of the corresponding wavelengths; the fitting value is calculated based on the polynomial function of each wavelength subinterval, for example, there is any wavelength value of the filtered spectrum data point , find the wavelength sub-interval corresponding to the wavelength value, and Substitute the polynomial function of the wavelength sub-interval for calculation, and the function value of the obtained polynomial function is the fitting value of the wavelength; based on the fitting value, the filtered spectrum data is updated to obtain updated accurate spectrum data.

[0062] The calculation formula of the polynomial function is: ;in, Indicates The polynomial function value corresponding to the wavelength value of any filtered spectrum data point in the wavelength sub-intervals; Indicates The wavelength value of any filtered spectrum data point in the wavelength sub-intervals; , , and are the polynomial coefficients.

[0063] The polynomial coefficients have a certain influence on the function value of the polynomial function. Randomly selecting certain constants from constants to assign values ​​to the polynomial coefficients may easily lead to errors in the function value of the polynomial function. Therefore, an objective function is constructed to solve the polynomial coefficients.

[0064] Construct an objective function, optimize the dynamic parameters of the objective function, and obtain the optimal dynamic parameters; calculate the optimal polynomial coefficient vector based on the optimal dynamic parameters; the elements in the optimal polynomial coefficient vector are the polynomial coefficients.

[0065] Objective Function ;in, Indicates the number of filtered spectrum data points in any wavelength subinterval in the filtered spectrum data; Indicates the number of The polynomial function value corresponding to the wavelength value of the filtered spectrum data point; express The corresponding wavelength sub-interval The light intensity value of each filtered spectral data point; Indicates a dynamic parameter.

[0066] Ways to optimize the dynamic parameters of the objective function include: The larger the value of the dynamic parameter, the easier it is to cause underfitting, and the spectral data cannot be fully fitted; the larger the value of the dynamic parameter, the easier it is to cause overfitting, and overfitting of the noise in the data will cause the changing trend of the fitted data to be incorrect. Therefore, it is necessary to select a suitable dynamic parameter to balance overfitting and underfitting.

[0067] Preset the initial value of the dynamic parameter, and calculate the dynamic norm based on the initial value of the dynamic parameter; continuously adjust the initial value of the dynamic parameter until the dynamic norm reaches the minimum value, and fix the dynamic parameter at this time as the optimal dynamic parameter.

[0068] Dynamic norm ;in, Represents the light intensity vector composed of the light intensity values ​​of all filtered spectral data points in the wavelength sub-interval corresponding to the objective function; represents a polynomial matrix; Indicates Dynamic parameters adjusted.

[0069] Polynomial Matrix ;in, Represents the wavelength value of the first filtered spectrum data point in the wavelength subinterval corresponding to the objective function; Indicates the first wavelength subinterval corresponding to the objective function The wavelength value of each filtered spectral data point.

[0070] Optimal polynomial coefficient vector ;in, represents the optimal dynamic parameter; represents an identity matrix; Represents the transpose of a polynomial matrix.

[0071] The calculation formula for updating the filtered spectrum data is: ; Represents any wavelength value in precise spectral data The corresponding light intensity value; Indicates wavelength value The corresponding fitted values; Indicates wavelength value The corresponding light intensity value in the filtered spectral data; Indicates wavelength value The maximum difference between the light intensity value of the wavelength sub-interval in the corresponding filtered spectrum data and the fitted value; Represents a constant ( ).

[0072] A multi-band and multi-purpose spectral filtering system is composed of a data acquisition module, a missing value processing module, an outlier detection module and a filtering module. The collected spectral data is processed by the system to obtain high-quality and high-precision spectral data. The system can be directly applied in experiments or process flows in other optical fields to improve work efficiency.

[0073] This embodiment collects spectral data, and uses a missing value processing module, an outlier detection module and a filtering module to finely process the collected spectral data, thereby realizing a multi-band and multi-purpose spectral filtering system. Compared with existing experience, the missing values ​​in the spectral data are processed by using an interpolation algorithm and a machine learning model, which not only improves the accuracy of the predicted value, but also can capture important features in the spectral data. At the same time, a custom scoring mechanism is introduced to calculate the loss function, so that the model can better adapt to the characteristics and laws of the spectral data. In terms of outlier processing, an efficient outlier judgment method is adopted. Compared with the traditional fixed threshold, the use of a dynamic threshold to judge outliers is more flexible and accurate. The spectral data is denoised by using a wavelet transform, and the spectral data is baseline corrected by using a cubic interpolation method, which effectively removes the background noise and interference signals in the spectral data, making the spectral data more accurate and indirectly improving the system work efficiency.

[0074] Embodiment 2; See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides a multi-band multi-purpose spectral filtering method, including: S1. Collect spectral data of experimental samples under a preset light source; S2. Process the missing values ​​of the spectral data to obtain complete spectral data; S3. Determine the outliers in the complete spectral data and process the outliers on the complete spectral data to obtain normal spectral data; S4. It includes a wavelet denoising unit and a baseline filtering unit. The wavelet denoising unit is used to perform denoising on normal spectral data to obtain filtered spectral data; the baseline filtering unit is used to perform baseline correction on the filtered spectral data to obtain accurate spectral data.

[0075] Embodiment 3; This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the multi-band multi-purpose spectral filtering method provided above is implemented.

[0076] Since the electronic device introduced in this embodiment is an electronic device used to implement a multi-band multi-purpose spectral filtering method in the embodiment of the present application, based on the multi-band multi-purpose spectral filtering method introduced in the embodiment of the present application, the technical personnel of the field can understand the specific implementation of the electronic device of the present embodiment and its various variations, so how the electronic device implements the method in the embodiment of the present application is not described in detail here. As long as the technical personnel of the field implement the electronic device used in the multi-band multi-purpose spectral filtering method in the embodiment of the present application, it belongs to the scope of protection of the present application.

[0077] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0078] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A multi-band multi-purpose spectral filtering system, characterized in that: include: A data acquisition module is used to collect spectral data of experimental samples under a preset light source; Missing value processing module, used to process missing values ​​of spectral data to obtain complete spectral data; An outlier detection module is used to determine outliers in the complete spectral data and perform outlier processing on the complete spectral data to obtain normal spectral data; The filtering module includes a wavelet denoising unit and a baseline filtering unit. The wavelet denoising unit is used to perform denoising on normal spectral data to obtain filtered spectral data; the baseline filtering unit is used to perform baseline correction on the filtered spectral data to obtain accurate spectral data; and each module is connected by wire and / or wireless means.

2. The multi-band multi-purpose spectral filtering system according to claim 1, characterized in that: Spectral data is collected using a preset light collection method; the spectral data includes wavelength, light intensity and corresponding timestamp.

3. The multi-band multi-purpose spectral filtering system according to claim 2, characterized in that: The method of processing missing values ​​of spectral data includes: The Monte Carlo algorithm is used to process the wavelength missing points of the spectral data, and the numerical values ​​generated by the algorithm are used to fill the wavelength missing values ​​of the spectral data. The effect of the wavelength missing value processing is evaluated by calculating the mean square error of the spectral data after the wavelength missing value processing; when the mean square error is less than the preset mean square error threshold, the wavelength missing value processing effect is optimal, and a wavelength filled data set is obtained at this time; Construct a machine learning model and use the RNN recurrent neural network model as the basic framework of the machine learning model; use the machine learning model to process the missing values ​​of light intensity in the wavelength filling dataset, and use the predicted values ​​of the machine learning model to fill the missing values ​​of light intensity in the wavelength filling dataset; Collect historical complete spectral data as training labels for machine learning models; randomly select data points, and mark the timestamps of these data points to obtain marked data points; delete the light intensity values ​​of all marked data points to obtain a missing light intensity data set, and use the missing light intensity data set as a training set for the machine learning model; Construct a custom scoring mechanism to calculate the custom score of each data point in the missing light intensity dataset and the average of the custom scores of all data points; define a machine learning loss function and calculate the function value of the machine learning loss function based on the custom score of each spectral data point and the average of the custom scores of all data points.

4. The multi-band multi-purpose spectral filtering system according to claim 3, characterized in that: The formula for calculating the custom score for each data point in the missing light intensity dataset is: ;in, Indicates the missing light intensity data set Custom scoring for each data point; Indicates the missing light intensity data set Physical law scores for each data point; represents the weight of the physical law score; Indicates the missing light intensity data set Smoothed scores of data points; represents the weight of the smoothing score; Indicates the missing light intensity data set missing scores for data points; represents the weight of missing ratings; Physics Law Rating ;in, Indicates the missing light intensity data set The predicted value of light intensity for each data point; Indicates the missing light intensity data set The true value of the light intensity at each data point; represents the standard deviation of light intensity; Smoothed Scoring ;in, Represents the total number of data points in the missing light intensity dataset; Missing rating ;in, Indicates the number of missing light intensity data in the preset The weight of the light intensity of each data point; Indicates the missing light intensity data set The light intensity value of each data point; Machine Learning Loss Functions ;in Represents the mean of the custom scores for all data points in the missing light intensity dataset; Training machine learning models using missing light intensity datasets; weighting physical law scores using gradient descent , weight of smoothing score and weights for missing ratings Update; when the function value of the machine learning loss function no longer decreases, fix the parameters to obtain a trained machine learning model.

5. The multi-band multi-purpose spectral filtering system according to claim 4, characterized in that: The method of processing outliers on the complete spectral data includes: Based on the timestamp of each complete spectral data point in the complete spectral data, the complete spectral data is sorted in chronological order to obtain ordered complete spectral data; the ordered complete spectral data is standardized to obtain standardized spectral data; characteristic coefficients are defined, and anomaly thresholds are calculated based on the characteristic coefficients; the standardized spectral data are processed using an anomaly detection algorithm to obtain an anomaly score for each standardized spectral data point in the standardized spectral data; all standardized spectral data points in the standardized spectral data whose anomaly scores are greater than the anomaly threshold are selected to form an anomaly data set; Calculate the information entropy of each abnormal spectral data point in the abnormal data set, calculate the weight vector of each abnormal spectral data point in the abnormal data set based on the information entropy of all abnormal spectral data points; calculate the abnormal factor score of each abnormal spectral data point in the abnormal data set; calculate the final abnormal value score based on the abnormal value score, weight vector and abnormal factor score of each abnormal spectral data point in the abnormal data set; if the final abnormal value score of any abnormal spectral data point in the abnormal data set is greater than the preset final abnormal value score threshold, the abnormal spectral data point is determined to be an error spectral data point; delete all error spectral data points to obtain normal spectral data.

6. The multi-band multi-purpose spectral filtering system according to claim 5, characterized in that: The method of calculating the abnormal threshold includes: Characteristic coefficient ;in, represents the number of standardized spectral data points in the standardized spectral data; Indicates The wavelength value of the standardized spectral data point; Represents the wavelength mean of all standardized spectral data points in the standardized spectral data; Indicates The light intensity value of the standardized spectral data point; Represents the mean light intensity of all standardized spectral data points in the standardized spectral data; Normalize the characteristic coefficients of all standardized spectral data points to obtain the characteristic coefficient vector ; The characteristic coefficient vector All characteristic coefficients in are arranged in descending order from large to small, and the top characteristic coefficients to form a characteristic coefficient set; summing all characteristic coefficients in the characteristic coefficient set, and calculating an abnormal threshold based on the sum of all characteristic coefficients in the characteristic coefficient set; Abnormal Threshold ;in, Represents the sum of all characteristic coefficients in the characteristic coefficient set; represents the regulating factor; The calculation formula of the weight vector is: ;in, Indicates abnormal spectral data points The weight vector of Represents the number of abnormal spectral data points in the abnormal data set; Indicates abnormal spectral data points Information entropy of represents an adaptive function; ;in, Indicates abnormal spectral data points The distribution probability of wavelength; Indicates abnormal spectral data points The distribution probability of light intensity; Adaptive function ;in represents an adaptive adjustment parameter; Indicates abnormal spectral data points The wavelength value of Indicates abnormal spectral data points The light intensity value; Represents the maximum value of wavelength among all abnormal spectral data points; Represents the maximum value of light intensity among all abnormal spectral data points.

7. The multi-band multi-purpose spectral filtering system according to claim 6, characterized in that: The method of calculating the abnormal factor score includes: For any two abnormal spectral data points in the abnormal data set, the Euclidean distance between them is calculated, and a distance matrix is ​​constructed based on the Euclidean distance between any two abnormal spectral data points; the smallest value is selected from each row of the distance matrix. The minimum Euclidean distance and sum of all abnormal spectral data points are summed to obtain the distance vector ; where the distance vector Any element in represents the minimum Euclidean distance sum of any abnormal spectrum data point; Represents the transpose of a vector; The inverse of the minimum Euclidean distance sum of any abnormal spectral data point is taken to obtain the abnormal factor score of the abnormal spectral data point; Final outlier score ;in, Indicates abnormal spectral data points outlier score; Indicates abnormal spectral data points The abnormal factor score of Represents a fusion weight.

8. The multi-band multi-purpose spectral filtering system according to claim 7, characterized in that: The method of performing denoising on normal spectral data includes: Preset wavelet basis function, from less than or equal to A positive integer is randomly selected from the positive integers as the decomposition layer number of the wavelet decomposition; the normal spectral data is subjected to wavelet decomposition based on a preset wavelet basis function to obtain detail coefficients; the detail coefficients are threshold processed based on a preset detail coefficient threshold to obtain denoised detail coefficients; the denoised detail coefficients are signal reconstructed using an inverse wavelet transform to obtain filtered spectral data.

9. The multi-band multi-purpose spectral filtering system according to claim 8, characterized in that: The method of performing baseline correction on the filtered spectral data includes: Selecting every five consecutive filtered spectrum data points from the filtered spectrum data as a group of correction points; taking the midpoint of any two adjacent correction points in a group of correction points, dividing the wavelength into three wavelength sub-intervals based on any two adjacent midpoints, constructing a polynomial function based on the wavelength sub-intervals; calculating fitting values ​​based on the polynomial function of each wavelength sub-interval; updating the filtered spectrum data based on the fitting values ​​to obtain updated accurate spectrum data; The calculation formula of the polynomial function is: ;in, Indicates The polynomial function value corresponding to the wavelength value of any filtered spectrum data point in the wavelength sub-intervals; Indicates The wavelength value of any filtered spectrum data point in the wavelength sub-intervals; , , and are the polynomial coefficients; Construct an objective function, optimize the dynamic parameters of the objective function, and obtain the optimal dynamic parameters; calculate the optimal polynomial coefficient vector based on the optimal dynamic parameters; the elements in the optimal polynomial coefficient vector are the polynomial coefficients; Objective Function ;in, Indicates the number of filtered spectrum data points in any wavelength subinterval in the filtered spectrum data; Indicates the number of The polynomial function value corresponding to the wavelength value of the filtered spectrum data point; express The corresponding wavelength sub-interval The light intensity value of each filtered spectral data point; Indicates a dynamic parameter.

10. The multi-band multi-purpose spectral filtering system according to claim 9, characterized in that: The method of optimizing the dynamic parameters of the objective function includes: Preset the initial value of the dynamic parameter, and calculate the dynamic norm based on the initial value of the dynamic parameter; continuously adjust the initial value of the dynamic parameter until the dynamic norm reaches the minimum value, and fix the dynamic parameter at this time as the optimal dynamic parameter; Dynamic norm ;in, Represents the light intensity vector composed of the light intensity values ​​of all filtered spectral data points in the wavelength sub-interval corresponding to the objective function; represents a polynomial matrix; Indicates Dynamic parameters of the secondary adjustment; Polynomial Matrix ;in, Represents the wavelength value of the first filtered spectrum data point in the wavelength subinterval corresponding to the objective function; Indicates the first wavelength subinterval corresponding to the objective function The wavelength value of the filtered spectral data point; Optimal polynomial coefficient vector ;in, represents the optimal dynamic parameters; represents an identity matrix; represents the transpose of a polynomial matrix; The calculation formula for updating the filtered spectrum data is: ; Represents any wavelength value in precise spectral data The corresponding light intensity value; Indicates wavelength value The corresponding fitted values; Indicates wavelength value The corresponding light intensity value in the filtered spectral data; Indicates wavelength value The maximum difference between the light intensity value of the wavelength sub-interval in the corresponding filtered spectrum data and the fitted value; Represents a constant.

Images